AI Writing Pattern Checker

Explore English writing patterns locally. The underlying RoBERTa detector was trained on GPT-2 outputs, not validated here for modern AI models. Load the model deliberately to begin; its score cannot establish authorship or publishability.

Text/settingsBrowser-only
Important:This is a probabilistic writing-pattern estimate, not proof of authorship. Do not use it as the sole basis for grading, hiring, discipline, or accusations.

Paste Your Text

This tool processes the data you enter in your browser during the analysis.

Words: 0Minimum 250 · Maximum 5,000 words
First load:This tool runs in your browser. The first load may take a moment while the model is cached.
Technical details
—WebGPU —WebAssembly —Web Workers —Cache Storage
Not loaded — click Load Detector to begin

Results appear here after analysis.

How AI Content Detection Works

AI detectors analyze writing patterns. Rather than claiming that individual words are AI-generated, this tool highlights paragraphs or sections whose combined model and writing-pattern signals are elevated.

What the detector examines

  • Sentence variation — Variation is descriptive, not evidence of who wrote a passage.
  • Word predictability — The legacy model estimates patterns learned from its training data.
  • Phrase repetition — Repeated phrases can suggest an editorial review, not an authorship conclusion.
  • Lexical diversity — Range of unique vocabulary provides another signal.
  • Section-level hotspots — Color coding shows where template markers, stock phrasing, or elevated model scores are concentrated.

Best Practices

Use as one signal

Pair the score with your own review.

Context matters

AI-pattern scores matter differently depending on content type.

Respect uncertainty

Trust mixed results rather than forcing a decision.

FAQ

Tool notes · Reviewed Jun 2026 · Sources and limitations

Review details: 2026-06-13 · Marc LaClear · v1.3-beta

Reference sources:

Known limits:

  • AI detection is probabilistic and can be wrong.
  • The deployed manifest identifies roberta-base-openai-detector (GPT-2 output detector), wrapper version 1.2, with writing-pattern overlay 1.1.0. The overlay has no published benchmark or calibration.
  • Color coding highlights paragraph- or section-level evidence only; it does not prove that specific words were generated.
  • Accuracy varies by text length, domain, author style, and AI model.
  • Edited AI text and non-native English may produce uncertain results.